w2v2_ablation_focal_ctc_a0.75_g0.5-best_on-ling_head-tp0.025_tl10_fp0.001_fl16

This model is a fine-tuned version of nguyenvulebinh/wav2vec2-base-vietnamese-250h on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 3.0151
• Wer: 0.0928

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 8
• eval_batch_size: 16
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• total_train_batch_size: 32
• total_eval_batch_size: 64
• optimizer: Adam with betas=(0.9,0.98) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1337.3806	0.94	100	875.6539	18.6402
928.5113	1.89	200	336.9626	17.0922
159.8738	2.83	300	65.9093	1.0
84.4363	3.77	400	60.3714	1.0
77.6119	4.72	500	57.3562	1.0
74.61	5.66	600	56.1581	1.0
73.6043	6.6	700	55.2834	1.0
72.9912	7.55	800	54.6511	1.0
71.2358	8.49	900	54.5294	1.0
69.8267	9.43	1000	52.0085	0.9654
58.2621	10.38	1100	28.9713	0.5667
32.9698	11.32	1200	12.7473	0.2391
20.9063	12.26	1300	8.4371	0.1719
15.8491	13.21	1400	6.8039	0.1471
13.5732	14.15	1500	5.7138	0.1358
11.2257	15.09	1600	5.0809	0.1287
10.3752	16.04	1700	4.7094	0.1257
9.4428	16.98	1800	4.4958	0.1240
8.7466	17.92	1900	4.2252	0.1121
8.8036	18.87	2000	4.0767	0.1063
7.8423	19.81	2100	4.0642	0.1169
7.7711	20.75	2200	3.7993	0.1062
7.3128	21.7	2300	3.6860	0.1018
7.0561	22.64	2400	3.5755	0.1031
7.0214	23.58	2500	3.5520	0.0994
6.3277	24.53	2600	3.5347	0.1000
6.4895	25.47	2700	3.4519	0.1090
6.0361	26.42	2800	3.4117	0.1075
5.6113	27.36	2900	3.4126	0.1044
5.4031	28.3	3000	3.4020	0.0991
5.408	29.25	3100	3.3015	0.0937
5.5346	30.19	3200	3.4108	0.0962
5.5502	31.13	3300	3.3039	0.0929
4.7607	32.08	3400	3.3695	0.1028
5.3438	33.02	3500	3.4625	0.1061
5.3239	33.96	3600	3.4865	0.1063
5.156	34.91	3700	3.3536	0.1001
4.7838	35.85	3800	3.3247	0.0999
4.7075	36.79	3900	3.2070	0.1022
4.8445	37.74	4000	3.1779	0.0963
4.7855	38.68	4100	3.2078	0.0973
4.3254	39.62	4200	3.2060	0.0984
4.4259	40.57	4300	3.2057	0.0967
4.4873	41.51	4400	3.0877	0.0931
4.6976	42.45	4500	3.0714	0.0963
4.0921	43.4	4600	3.0722	0.0888
3.6267	44.34	4700	3.1064	0.0943
3.8833	45.28	4800	3.0917	0.0874
3.8643	46.23	4900	3.1006	0.0881
3.7386	47.17	5000	3.0927	0.0873
3.4363	48.11	5100	3.0982	0.0891
3.5792	49.06	5200	3.0596	0.0906
3.3444	50.0	5300	3.0289	0.0951
3.3686	50.94	5400	3.0119	0.0858
3.6072	51.89	5500	3.0416	0.0986
3.7266	52.83	5600	3.0389	0.0950
3.6465	53.77	5700	3.0102	0.0945
3.2426	54.72	5800	3.0769	0.1012
3.1878	55.66	5900	2.9749	0.0956
3.1891	56.6	6000	3.0639	0.0912
3.2342	57.55	6100	3.0031	0.0958
2.9652	58.49	6200	2.9965	0.0993
3.1089	59.43	6300	3.0358	0.0914
2.9434	60.38	6400	3.0805	0.0948
3.2816	61.32	6500	3.0516	0.0944
3.1317	62.26	6600	3.0206	0.0902
3.1278	63.21	6700	3.0254	0.0973
3.1522	64.15	6800	3.0528	0.0970
3.0941	65.09	6900	3.0627	0.0970
3.1021	66.04	7000	3.0484	0.0992
2.8751	66.98	7100	3.0559	0.0953
2.8807	67.92	7200	3.0577	0.0982
3.2996	68.87	7300	3.0628	0.0944
2.9746	69.81	7400	3.0304	0.0948
2.7453	70.75	7500	3.0483	0.0936
2.7083	71.7	7600	3.0759	0.0958
2.531	72.64	7700	3.0622	0.0962
2.6315	73.58	7800	3.0232	0.0921
2.4475	74.53	7900	3.0046	0.0918
2.6836	75.47	8000	3.0124	0.0924
2.7316	76.42	8100	3.0200	0.0896
2.7433	77.36	8200	3.0580	0.0936
2.5052	78.3	8300	3.0516	0.0934
3.1428	79.25	8400	3.0461	0.0936
2.7542	80.19	8500	3.0198	0.0947
2.7269	81.13	8600	3.0262	0.0945
2.7809	82.08	8700	3.0139	0.0897
2.3545	83.02	8800	3.0183	0.0924
2.4138	83.96	8900	3.0209	0.0921
2.4908	84.91	9000	3.0268	0.0924
2.6911	85.85	9100	3.0228	0.0948
2.4881	86.79	9200	3.0194	0.0922
2.6499	87.74	9300	3.0090	0.0908
2.5886	88.68	9400	3.0162	0.0917
2.6444	89.62	9500	3.0180	0.0909
2.5907	90.57	9600	3.0199	0.0908
2.6175	91.51	9700	3.0198	0.0923
2.8366	92.45	9800	3.0164	0.0915
2.5604	93.4	9900	3.0118	0.0912
2.4371	94.34	10000	3.0124	0.0908
2.6646	95.28	10100	3.0187	0.0920
2.5563	96.23	10200	3.0140	0.0919
2.8501	97.17	10300	3.0144	0.0919
2.6802	98.11	10400	3.0150	0.0923
2.3091	99.06	10500	3.0150	0.0926
2.6642	100.0	10600	3.0151	0.0928

Framework versions

• Transformers 4.35.2
• Pytorch 1.13.1+cu117
• Datasets 2.12.0
• Tokenizers 0.14.1